Curr Psychol DOI 10.1007/s12144-014-9224-7
A Preliminary Investigation into Effects of Linguistic Abstraction on the Perception of Gender in Spoken Language A. B. Siegling & Michelle Eskritt & Mary E. Delaney
# The Author(s) 2014. This article is published with open access at Springerlink.com
Abstract We investigated the role that linguistic abstraction may play in people’s perceptions of gender in spoken language. In the first experiment, participants told stories about their best friend and romantic partner. Variations in linguistic abstraction and gender-linked adjectives for describing their close others were examined. Participants used significantly more abstract language to describe men compared to women, possibly reflecting a gender stereotype associated with the dispositionality factor of linguistic abstraction. In a second experiment, a new group of participants judged the gender of the protagonists from the stories generated in Experiment 1, after the explicit linguistic gender cues were removed. Consistent with the dispositionality factor, linguistic abstraction moderated the effects of the gender stereotypicality of the context (masculine, feminine, or neutral) on participants’ gender judgments. Discussion focuses on the implications of the results for the communication of gender stereotypes and the effects of linguistic abstraction in more naturalistic language. Keywords Gender . Spoken language . Stereotypes . Linguistic abstraction . Linguistic category model Although Western culture seeks to become gender-egalitarian and societal disapproval of gender stereotypes is becoming more widespread, progress has been slow. More subtle forms of sexist practice, in particular, still persist (Holmes 2005) and many people continue to express gender stereotypes in implicit ways, which has been referred to as modern sexism (Swim et al. 1995). Language, especially interpersonal discourse, is one medium through which gender stereotypes are transmitted implicitly
This research was supported in part by an internal university grant to the second and third authors. A. B. Siegling (*) Division of Psychology and Language Sciences, University College London, WC1H 0AP London, UK e-mail:
[email protected] M. Eskritt : M. E. Delaney Mount Saint Vincent University, Halifax, Canada
Curr Psychol
(e.g., Harasty 1997; Lenton et al. 2009; Ochs 1992). However, investigations of how gender is represented in our language and, perhaps more importantly, how such language-mediated gender distinctions are perceived by recipients have largely concentrated on the use of explicitly gendered terms, which tend to be overtly sexist. Examples include the generic pronoun he to refer to any person, using job titles ending in man, or the asymmetry of titles as in using Mrs. or Miss to indicate women’s marital status (Speer 2002). Consistent with the Sapir-Whorf hypothesis that language can influence thought (e.g., Whorf 1956), evidence suggests that the use of explicitly gendered terms is related to how we process and perceive information. For example, the use of the generic words he and man was found to elicit gender-biased thinking, specifically male imagery, in children and adults (Henley 1989). Similarly, American undergraduates were more likely to think of female characters when reading sentences involving he/she or they than for sentences containing the generic he (Khosroshani 1989). Ochs (1992) argued that in the English language there are few such direct indices of gender (beyond he/she, Mr./Miss/Mrs., etc.) and instead gender is frequently indexed indirectly. For example, some occupations are associated with a particular gender regardless of whether or not the occupation is explicitly gender-linked by adding man to the end of the title (Lassonde and O’Brien 2013; Reynolds et al. 2006). Gender enactment is multifaceted and more subtle, going beyond simple practices like the use of gender-inclusive language to include practices such as drawing attention to gender or one particular gender either directly or indirectly (Hopper and LeBaron 1998). For example, Sunderland (2002) examined a selection of parenting manuals to see how fathers were portrayed. She noted that fathers played an almost optional role in parenting with little reference to them and, when they were discussed, they were frequently portrayed as “bumbling assistants,” “line managers,” and/or “baby entertainers.” However, beyond semantic content as an indirect way of indicating gender, it is less clear whether there are specific word categories (like personal pronouns) that are not explicitly gender-linked, but nonetheless used to indicate gender implicitly. Language of this kind may also influence recipients’ perception of information in a gender-biased fashion.
Linguistic Categories, Abstraction, and Cognitive Inferences Interpersonal verbs and adjectives comprise the most commonly used linguistic categories in person and behavior descriptions (Semin and Fiedler 1988). One of these categories, adjectives, encompasses a few words that are explicitly gender-linked (e.g., masculine, feminine; male, female) as well as a wider range of implicit, albeit stereotypically gender-linked terms. Implicitly gender-linked adjectives reflect traditional gender-role norms and personality traits on which men and women are expected to differ (Bem 1974; Williams and Bennett 1975). For example, men are stereotypically assumed to be strong and aggressive, whereas women are thought to be more emotional and passive. Interpersonal linguistic categories vary on a concreteness–abstractness dimension, ranging from neutral descriptions of concrete actions (e.g., to kiss, to stare) to psychological states (e.g., to like, to envy) to more abstract qualities, such as traits or
Curr Psychol
dispositions (e.g., reliable, aggressive; see Semin and Fiedler 1988). The use of interpersonal verbs and adjectives in person descriptions has social-cognitive implications, as people make different attribute inferences based on the level of linguistic abstraction (e.g., Brown and Fish 1983; Fiedler and Semin 1988). Linguistic abstraction can reflect communicators’ thinking, and contribute to recipients’ inferences in two different ways, indicating dispositionality and event instigation (Semin and Marsman 1994). Research regarding dispositionality indicates that individuals described in more abstract interpersonal language are seen as stable in that characteristic, regardless of context (Semin and Fiedler 1988). For example, describing someone as kind is typically interpreted as a trait that will persist and reveal itself in future situations. In contrast, the use of more concrete interpersonal verbs shifts the level of explanation from enduring personal or group characteristics to contextual features in accounting for particular behaviors. If someone is described as having given money to a charity, the likelihood of this behavior occurring again is seen as less predictable compared to a generous supporter of charities. Although research in this area has focused on the behavior of individuals rather than groups, it would seem logical that the same principle (i.e., of characteristic behavior being described more abstractly) applies to the behavior of groups. Thus, if people perceive one gender’s behavior as more stable or predictable than the other gender’s behavior (i.e., due to a gender stereotype), then this may be reflected in a direct association between linguistic abstraction and gender in person descriptions and perceptions. The second type of inference, event instigation, focuses on who initiated the event being described. Specifically, the use of concrete action verbs (e.g., X compliments Y) also tends to be interpreted by listeners as the actor or sentence subject initiating events, whereas the use of the more abstract state verbs (e.g., X likes Y), which describe feeling states, tends to make the object in the sentence the causal agent (e.g. Brown and Fish 1983; Fiedler and Semin 1988). Thus, concrete action verbs assign the protagonist an active role, whereas abstract state verbs depict the protagonist as being more passive. This type of inference may also influence people’s use of linguistic abstraction in describing groups, such as men and women; a gender stereotype that one gender is more passive than the other gender could lead people to describe individuals of this gender in more abstract language. Research aiming to demonstrate a direct relationship between gender and linguistic abstraction, reflecting a gender stereotype implicit in either of the two factors of linguistic abstraction, would need to demonstrate that this link is independent of the more overtly gender-linked adjectives. A third mechanism through which linguistic abstraction may be linked to gender is as a function of men’s and women’s actual behavior, relative to the behavior one can expect from them based on gender-role norms. Maass et al. (1995) demonstrated that both individual and group behavior that is consistent with expectations or stereotypes is described more abstractly than behavior that violates expectations. This phenomenon has been referred to as the linguistic expectancy bias (LEB; Wigboldus et al. 2000) and can be attributed to the greater stability and predictability of expected information. In consideration of the LEB, linguistic abstraction may also vary with gender to reflect men’s and women’s relative conformity to gender-role norms. One gender may be more likely to conform to gender-role norms than the other, leading people to describe members of this group in more stereotypical contexts and, hence, use more abstract language. This possibility implies that linguistic abstraction varies with individual
Curr Psychol
differences in conformity to gender-role norms, with conforming individuals being described more abstractly than less conforming persons.
Effects of Linguistic Abstraction on the Recipient Building on the above findings, Wigboldus et al. (2000) asked participants to provide examples of situations where a male and a female friend acted in either a stereotypically consistent or stereotypically inconsistent manner, relative to gender-role norms. Recipients reading the gender-stereotypical descriptions were more likely to attribute the behavior to stable1 and enduring traits than stereotypically-inconsistent descriptions, which they were more likely to attribute to situational factors. Importantly, linguistic abstraction was found to mediate the effect of expectation consistency (i.e., relative to group-level stereotypes) on recipients’ dispositional attributions; recipients’ inferences about the person described varied as a function of the level of linguistic abstraction used by the communicator, with more abstract language reflecting greater stereotype consistency of descriptions and increasing the likelihood of dispositional attributions. However, this investigation did not fully control the effects of shared cultural stereotypes, as recipients knew the stereotypes implicit in the messages they were reading. That is, speakers were asked to describe events in which either a male or female friend acted in either a stereotypically masculine or feminine way. In a later study, the same researchers observed that linguistic abstraction influenced inferences about individuals (i.e., their personalities), who were unknown to the recipients, in the same way (Wigboldus et al. 2006). Transmitters were asked to describe an event where a friend acted expectedly and a scenario where this friend acted unexpectedly, relative to the friend’s personality. Communicators described behaviors of their friends in expected events in more abstract language than behaviors in unexpected events. Again, recipients were more likely to attribute behavior to the person’s personality when the behavior was described in abstract language, whereas behavior described in more concrete interpersonal language was judged to be influenced more by the situation; linguistic abstraction mediated the effect of the level of expectancy on dispositional attributions. Together, these studies provided strong evidence that linguistic abstraction can influence people’s perception of information. A strength of the research by Wigboldus et al. (2000, 2006) was the reliance on participants’ own written language, which lends a more naturalistic quality to the data and, thus, enhances the potential real-world relevance of the findings. Prior research on the role of linguistic abstraction in communicating about groups focused on participants’ interpretation of experimenter-produced written language, usually in the form of simple, isolated sentences (e.g., Maass et al. 1995; Semin and Fiedler 1988). While this is a valuable direction to pursue, participants were cued for role expectations in both studies, being asked explicitly to describe others in expected and unexpected ways (Wigboldus et al. 2000, 2006). This procedure, although useful in demonstrating LEB effects on recipients, does not reveal much about how linguistic abstraction is used in 1 It is important to emphasize that the word ‘stable’ or ‘stability’ is used in the literature to refer to the consistency and, thus, predictability of behavior, rather than to denote its quality (as in lability or saneness).
Curr Psychol
natural language without any directions for describing others imposed on the speaker. Further, it is unclear how linguistic abstraction as it appears in natural language might influence the recipient in making inferences about the protagonist. In natural language, it may not be apparent relative to which general attribute (gender, ethnicity, social class, personality, etc.) linguistic abstraction is used to indicate how expected the information communicated is for the person being described. Linguistic abstraction could be concrete because the protagonist did not conform to gender-role norms, act in line with his or her personality, or behave according to cultural customs in the event described. For various reasons (e.g., instructions, biases, and the context of information presented), the recipient may then erroneously use linguistic abstraction to make inferences about some other general attribute of the person—one that is unrelated to linguistic abstraction, because the speaker used a different attribute as the reference point. When the recipient assumes an inappropriate reference point for linguistic abstraction, as is possible in natural language, LEB effects may not be “picked up” by the recipient in a clear-cut way. To our knowledge, no study has examined the effects of linguistic abstraction in more natural language without requesting, and, thus, cuing for particular norm-referenced descriptions, such as gender stereotypes. Further, spoken language is more spontaneous, less carefully considered, and characterized by a smaller degree of complexity in terms of structure than written language (Miller and Weinart 2009).
Present Study The present study was designed to explore the potential role of linguistic abstraction in people’s perceptions of gender in fairly naturalistic language. Although some words through which linguistic abstraction manifests itself are stereotypically, and sometimes even explicitly gender-linked, the abstraction dimension also underlying the use of interpersonal verbs and adjectives would represent a more subtle way of indicating gender. Simultaneously, the study was intended to provide some insight into the more general effects of linguistic abstraction in natural language, particularly as a potential means of conveying stereotypic content in interpersonal discourse. Whereas linguistic abstraction was previously found to explain (i.e., mediate) much of the effect of expectation conformity on recipients’ dispositional attributions about groups or individuals, it may also alter, or moderate, the effect of information pertaining to other attributes not used as a reference point for linguistic abstraction on recipients’ inferences. As discussed, linguistic abstraction may also transmit meaning directly, considering the two distinct types of inference (i.e., dispositionality and event instigation) it can elicit. In contrast to previous research into the effects of linguistic abstraction, more natural, spoken language elicited in Experiment 1 provided the basis for text stimuli to be interpreted in Experiment 2. We asked speakers to tell stories about their romantic partner and best friend, expecting that in most cases one of them would be a woman and the other a man (see the Appendix for an example of a story told about a female and male protagonist). Stories were requested in a manner that identified individuals by their relationship to minimize the salience of gender and hopefully avoid cueing gender stereotypes. Instead of asking speakers to tell gender-stereotypical or atypical stories,
Curr Psychol
the stories were coded retrospectively based on their consistency with gender-role norms. Participants in Experiment 2 read transcripts of these stories and judged the gender of the protagonist with explicit linguistic gender cues removed. These gender judgments acted as a more specific measure of dispositionality, or personversus-situation attributions. Through regression analysis, we were then able to examine the role of linguistic abstraction in recipients’ perception of the protagonists’ gender.
Experiment 1 Experiment 1 was designed to elicit spoken language by participants about female and male individuals to be used as stimuli in Experiment 2. Participants were asked to tell stories about their romantic partner and best friend. A secondary aim of the first experiment was to explore these stories for linguistic gender indices, focusing on the linguistic abstraction of interpersonal verbs and adjectives, which could influence recipients’ perceptions of the protagonists in Experiment 2. Consistent with the possibilities discussed in the introduction, we examined if the LEB or either the dimension of dispositionality or event instigation may influence participants’ use of linguistic abstraction systematically between descriptions of female and male close others. In particular, we wanted to explore whether or not linguistic abstraction would vary depending on the gender of the close other and/or the level of gender stereotypicality of the event described. For this purpose, close-other depictions were classified into stereotypical, neutral, and atypical, relative to masculine or feminine gender-role norms and depending on the close other’s gender. If the LEB leads participants to describe close others of one gender in more abstract language than the other, this effect should be mediated by the context’s level of gender stereotypicality (i.e., masculine, feminine, or neutral). Precisely, the effect of closeother gender on linguistic abstraction should be non-significant, or at least weaker, after controlling for the stories’ level of gender stereotypicality. The LEB holds that linguistic abstraction varies depending on how stereotypical the information communicated is for the group or person being described. As discussed, stereotypical information is described more abstractly than atypical information. Differences in the extent to which men and women are described in gender-stereotypical ways could, therefore, elicit systematic differences in the level of abstraction used in descriptions between these two groups. A direct effect of close-other gender on linguistic abstraction, unmediated by the level of gender stereotypicality, would suggest that the LEB does not explain this effect. Rather, it would be indicative of a gender stereotype implicit in either of the two inferences linked to linguistic abstraction: dispositionality or event instigation. Gender-linked adjectives may also influence the inferences made by the recipients in Experiment 2 about the gender of the person described. Since they fall into one of the categories of linguistic abstraction (the most abstract one), they represent a potential confound in regards to the aims of the experiment, given the focus on gender. Therefore, we also examined the natural occurrence of gender-linked adjectives in the close-other descriptions in this experiment, categorizing them based on their association with traditional gender stereotypes and gender-linked personality traits.
Curr Psychol
Method Participants Twenty-one undergraduate and seven graduate students, ranging in age between 18 to 41 years, were recruited from various disciplines at two eastern Canadian universities. The majority of participants (82 %) were between 21 and 29 years old. Eighteen of the participants were female and the other ten were male. Both graduate (2 male, 5 female) and undergraduate students (8 male, 13 female) were recruited to provide a greater range of exposure to egalitarian values, which can be expected to vary with the level of education. Even though such education-based exposure may influence the amount of overt sexist language, it is less clear that it will influence more subtle gender distinctions in language. As an incentive, participants’ names were entered into a drawing for a gift certificate. All participants were in a close relationship for at least one year with both their best friend and romantic partner, who had to be of different sexes. Data from two additional participants (one female, one male) were not included, as their romantic partners and best friends were of the same sex and, therefore, their use of language to describe female and male close others could not be compared. One male participant’s romantic partner was of the same sex but his best friend was of the other sex. His data were included in the results reported as they met the inclusion criterion of having a best friend and a romantic partner of different sexes. Also, including this case did not affect the observed patterns and significance of p levels. The length of relationships ranged from 1 to 23 years, with a mean length of 4.79 years (SD=3.17) for romantic partners and 11.26 years (SD=5.59) for best friends. Procedure Participants were asked to tell stories about their best friend and romantic partner (as opposed to a female and male close other), thereby eliciting one story of the other sex and one of the same sex. A female experimenter interviewed each participant individually.2 Before the interview, participants were informed that they would be asked to tell a story about their best friend and romantic partner that is a good example of the type of person they are. Hence, they had time to think about appropriate stories to tell. The order in which participants talked about their romantic partner and their best friend was counterbalanced by gender. Half of the participants of each gender talked about their romantic partner first, whereas the other half talked about their best friend first. The experimenter guided each participant through the interview by asking the following questions, first about the romantic partner or the best friend and then the same questions about the other person: Question 1: “How long have you known your partner/best friend?” Question 2: “Tell me what your best friend/partner is like. List a few of their qualities.” Question 3: “How did you meet your partner/best friend?” Question 4: “Tell me a story about your romantic partner/best friend that is a good example of the type person they are.” If necessary: “Can you think of a particular instance as an example of (giving one of the qualities listed in Question 2)?” 2 We specifically chose to use a female interviewer, as both men and women feel more comfortable disclosing personal information, especially emotion-related information, to a female than male recipient (Snell et al. 1989).
Curr Psychol
The purpose of Question 2 was to warm up each participant, to elicit some adjectives for later comparison, and to provide qualities to prompt the participant if necessary in Question 4. Question 3 was asked to elicit a more gender-neutral story, whereas Question 4 was more open-ended. The session took approximately 20 min was videotaped for later transcription. Scoring. Two independent coders (one female, one male) scored all transcriptions of the interviews. Interpersonal verbs and adjectives pertaining to the close other of the story only (i.e., the romantic partner or best friend) were identified. To look at the more overt referencing of gender, total numbers of the different adjective types (i.e., feminine, masculine, and neutral) were derived from the answers to both the stories and the request to list qualities of the close others (i.e., Questions 2, 3, and 4). For the more subtle references to gender, a linguistic abstraction score was obtained from the stories told about the best friend and romantic partner only (i.e., Question 3 and 4). Frequently, more than one story was told for Question 4 and even sometimes for Question 3, in which case all stories were examined and included within the different linguistic aggregates. To be able to examine linguistic abstraction in relation to the level of gender stereotypicality, separate abstraction scores were also derived for each story. Only verbs and adjectives associated with a close other were included in the scoring where the close other was a sentence subject. Participants narratives consistently depicted close others as the sentence subject. Gender-linked adjectives Adjectives were classified as either masculine (e.g., aggressive, dominant), feminine (e.g., affectionate, emotional), or neutral (e.g., happy, jealous). Available classifications of gender-linked adjectives (Bem 1974; Williams and Bennett 1975) facilitated this part of the coding procedure, though the lists were not exhaustive for the adjectives encountered. Nevertheless, inter-rater agreement between the male and female rater was perfect (κ=1.00). Totals of the three adjective types were calculated. Linguistic abstraction The linguistic category model (Semin and Fiedler 1988, 1991) classifies interpersonal verbs and adjectives into categories, with increasing levels of abstraction: descriptive action verbs (e.g., hit, yell, walk), interpretive action verbs (e.g., help, tease, avoid), state action verbs (e.g., surprise, amaze, anger), state verbs (e.g., admire, hate, appreciate), and adjectives (e.g., honest, reliable, aggressive). Thus, descriptive action verbs represent the most concrete word category, and adjectives represent the most abstract word category. Consistent with the “Linguistic Category Model Manual” (Coenen et al. 2006), each category was weighted by assigning a numerical value to it. Descriptive action verbs were assigned a score of 1, state verbs by a score of 3, and adjectives by a score of 4. Interpretive action verbs and state action verbs represent the same level of abstraction and were thus collapsed in one category, which was denoted by a score of 2. The inter-rater reliability was near perfect (κ=0.98). An average abstraction score is obtained by adding all scores and dividing the sum by the number of coded items, giving a possible range of 1 (very concrete descriptions) to 4 (very abstract descriptions). Level of gender stereotypicality All stories were rated in terms of how stereotypical (n=47), neutral (n=60), and atypical (n=10) they are, relative to the gender-role norms pertaining to the close other’s gender. Again, the available classifications of gender-
Curr Psychol
linked adjectives (Bem 1974; Williams and Bennett 1975) were used to help facilitate judgment and provide a common reference point for the two raters. The raters were asked to judge the story as a whole in regards to the gender stereotypicality of the close other’s behavior. An equal number of male and female close others were described in gender-neutral and atypical contexts. Of stories classified as stereotypical, 27 described a male close other. Again, inter-rater agreement was high (κ=0.95).
Results The length of the acquaintanceship with the romantic partner and best friend was not related to any of the dependent measures: totals of masculine, neutral, and feminine adjectives, as well as linguistic abstraction scores (p>0.05). Therefore, this variable was not included in further analyses. As the number of graduate students recruited was small, data were also collapsed across level of education. Although both women and men share stereotypical beliefs about gender-role behavior (Williams 1982), participant gender was included in the analyses to control for any gender-of-speaker effects. Gender-linked adjectives Table 1 displays the means and standard deviations for the different linguistic measures across narratives of female and male close others by female and male speakers. A 2 (speaker gender) x 2 (close-other gender) x 3 (adjective type) mixed-design ANOVA compared participants’ use of masculine, feminine, and neutral adjectives to describe their female and male close other. Mauchly’s test indicated
Table 1 Experiment 1: Means and Standard Deviations for Totals of Adjective Types and Linguistic Abstraction Scores as a Function of Close-Other Gender and Speaker Gender (N=28) Close-other gender
Male speakers
Female speakers
Total
M
SD
M
SD
M
SD
Male
1.60
2.50
1.28
1.23
1.39
1.75
Female
1.30
1.57
0.50
0.79
0.79
0.17
Male
0.50
0.71
1.22
1.63
0.96
1.40
Female
1.20
0.92
0.72
1.18
0.89
1.10
Male
7.90
5.26
9.00
5.12
8.61
5.10
Female
6.70
4.08
6.56
3.11
6.61
3.41
Male
2.69
0.36
2.51
0.38
2.57
0.37
Female
2.38
0.41
2.39
0.49
2.38
0.45
Adjective types Masculine adjectives
Feminine adjectives
Neutral adjectives
Linguistic abstraction
Linguistic abstraction scores have a possible range of 1 (very concrete descriptions) to 4 (very abstract descriptions)
Curr Psychol
violation of sphericity, χ2(2)=26.05, p
